On Wed, Aug 13, 2014 at 01:50:24PM -0400, Rik van Riel wrote: > On 08/13/2014 01:40 PM, Peter Zijlstra wrote: > > > So the problem with the above is the lack of fwd progress; if > > there's enough clone()/exit() happening in the thread group (and > > the more CPUs the more possible), we'll keep repeating. > > We can fall back to taking the lock if we circle around, > or if there is a writer active when we are in seqcount_read, > similar to what the semaphore (ipc/sem.c) code is doing. > > read_seqbegin_or_lock would do the trick...
Yep that would work.
pgp6SUyRnbmtg.pgp
Description: PGP signature
